Research Of Human Behavior Recognition Based On Kinect

With the success of the motion capture equipment launched and popularized, theyhave been applied in the field of intelligent surveillance, human-computer interaction,human behavior recognition etc. The study of human behavior representation methodand recognition method based on the data obtained from Kinect.In order to understand the human behavior better, According to the characteristicsof the human structure, this paper is based on3D human joint data obtained from Kinectto extract the angle information and the relative distance factors from the human jointvectors,as the characteristic parameters to represents the human posture.In this paper, the contents of human action recognition is divided into two parts,contain a typical static posture of the human behavior recognition and does not containa typical static posture of the human dynamic action recognition. Through theestablishment of the static maintain model of human posture to determine whetherhuman action sequence has contains typical static posture, then using the improveddecision tree classification algorithm to identification the typical static posture of thehuman behavior.According to the characteristics of DTW algorithm and KNN algorithm,the paper proposed a new parameters, the category of credibility, to compare thesimilarity of the human motion sequence is does not contain a typical static posture. Theparameters applied to the classification algorithm that has complete the identification ofthe dynamic action that does not contain a typical static posture.The experimental results show that the aimed at contain a typical static posture ofthe human behavior and does not contain a typical static posture of the human dynamicaction are proposed method for human behavior recognition.These method can achievegood test results and has a higher recognition accuracy than the existing methods.